Keeping control loop subsystems in healthy conditions is crucial for extended life of space systems, however, this cannot be always the case in that harsh environment. Therefore, forecasting the health status of sensors and actuators improves operations management such as anticipating the soft-update of some parameters which can be time consuming. This paper proposes an integrated solution to enhance the performance of Long-Short Term Memory (LSTM) network in predicting the remaining useful life (RUL) of angular rate gyros. This can be done by leveraging the eigen signature of gyro drift time series. To overcome the delays of baseline RUL prediction, our algorithm assesses the effect of sensor drift on eigen behavior of the sliding transformation matrix. The model input is then augmented to provide the LSTM with more key features for better performance. The proposed solution shows better performance in terms of prediction accuracy and optimizing the deep learning model by reducing the number of learnable parameters. Further, compared to more accurate methods, the proposed approach achieves a favorable balance between prediction performance and computational efficiency, requiring fewer resources while maintaining competitive accuracy.
Henna et al. (Thu,) studied this question.